parallel processing experiment

[scott petito]

I was experimenting today with plugins for parallel compression and eq for ITB...
what started out as a way to verify delay compensation turned into an interesting study of phase shift

I sent my 10 tracks of drums to 2 stereo auxes bus1-2 and bus 3-4
I made sure all levels were Identical to the sends by sending automation data to sends

then I put a trim plug on both aux tracks , I flipped the phase one aux bus trim bus1-2
I then paned both aux to center (mono)

the audio then cancelled 100% as it should.... I then decided to test delay compensation for Compression

as i inserted various compression plugs on aux 2 feed by 3-4 I noticed they all produced various phase shifts...remember the trims are still out of phase
even at unity gain with no compression some compressors "leaked audio" more than others

I tried tdm and rtas versions of the plugs some like most of the waves plugs sounded identical

the BF actually sounded quite different in RTAS and TDM

A UAD Fairchild was especially "leaky" at what I believed was unity settings...

remember I'm still only listening to the sum of the 2 out of phase buses

here's where it got interesting I tried to get a compression setting engaged and adjust it to produce as little leakage as possible( which obviously is phase shift caused by the change in level and attack release times) basically I was using attack and release times and some level compensation to get as close to phase cancelation as I could...

I then flipped the phase on the first bus... thus bringing all the drums back in to the mix including the parallel buss compression

it sounded really good....
here s the rub when I tried to increase the parallel buss fader to get more compression I became immediately aware of the phase shift ... it was subtle but obvious...

I tried this with various compressors at similar attack and release times and some compressors produced much more pleasant but still obvious phase shift
I started to wonder if this was due in part to the coloration of various compressors

the UAD Fairchild seemed to be the most forgiving and the dyn 3 was pretty good
others were hit and miss... strangely enough the linear multibands i used were the least forgiving
I did this with various eq's as well very interesting results there as well

well to end a long winded story .. I will be checking my parallel compression in mono and out of phase....

anyone else try this? ... I'm going to give it go with outboard but I already have convertor delay to cope with...


cheers
SP

[Berolzheimer]

Do you have delay compensation on? You should try having the exact same inserts on both auxes, with the plugs on one of them bypassed, to make sure the timing is correct. Sometimes it seems that plugs don't report the correct latency to PT, so the compensation delay isn't correct. That said a lot of plugs, especially ones that incorporate FFR filters or modelling of analog devices, do induce some phase shift.

[scott petito]

yes I had Delay compensation on... I was actually testing it by nullin the buses out of phase

the curious thing is I always thought having the same plug on both busses was a good thing to do but in mon with the busses out of phase having the same plug on both was even worse in terms of percieved phase shift

cheers
SP

[DannyMac]

I find para comp to sound best in the analouge domain. I've tried all DAW's and a few DIGI desks, but still sounds better on a cheapo desk

[thethrillfactor]

Originally Posted by scott petito

I was experimenting today with plugins for parallel compression and eq for ITB...
what started out as a way to verify delay compensation turned into an interesting study of phase shift

I sent my 10 tracks of drums to 2 stereo auxes bus1-2 and bus 3-4
I made sure all levels were Identical to the sends by sending automation data to sends

then I put a trim plug on both aux tracks , I flipped the phase one aux bus trim bus1-2
I then paned both aux to center (mono)

the audio then cancelled 100% as it should.... I then decided to test delay compensation for Compression

as i inserted various compression plugs on aux 2 feed by 3-4 I noticed they all produced various phase shifts...remember the trims are still out of phase
even at unity gain with no compression some compressors "leaked audio" more than others

I tried tdm and rtas versions of the plugs some like most of the waves plugs sounded identical

the BF actually sounded quite different in RTAS and TDM

A UAD Fairchild was especially "leaky" at what I believed was unity settings...

remember I'm still only listening to the sum of the 2 out of phase buses

here's where it got interesting I tried to get a compression setting engaged and adjust it to produce as little leakage as possible( which obviously is phase shift caused by the change in level and attack release times) basically I was using attack and release times and some level compensation to get as close to phase cancelation as I could...

I then flipped the phase on the first bus... thus bringing all the drums back in to the mix including the parallel buss compression

it sounded really good....
here s the rub when I tried to increase the parallel buss fader to get more compression I became immediately aware of the phase shift ... it was subtle but obvious...

I tried this with various compressors at similar attack and release times and some compressors produced much more pleasant but still obvious phase shift
I started to wonder if this was due in part to the coloration of various compressors

the UAD Fairchild seemed to be the most forgiving and the dyn 3 was pretty good
others were hit and miss... strangely enough the linear multibands i used were the least forgiving
I did this with various eq's as well very interesting results there as well

well to end a long winded story .. I will be checking my parallel compression in mono and out of phase....

anyone else try this? ... I'm going to give it go with outboard but I already have convertor delay to cope with...


cheers
SP

Well if you try it with a real analog compressor like a Fairchild it will not null either. The plug ins are designed to change the signal similar to what analog gear does even when not doing any processing.

I think the only one's that null are the Digi stock ones.

Also with RTAS plugs i think you are supposed to put them first in the chain.

[Kiwiburger]

Agreed. The cheap or freebie compressors often don't show any phaseshifting when null tested - because they aren't attempting to color the sound. Assuming you stay under the threshold.

The better sounding plugins always have some coloration built in, and often the phase shifting is a deliberate attempt to sound "analog".

The trick is to make the phase shifting work for you ... like "spill", problematic spill can become desirabe "ambiance" if you embrace it and work with it rather than try to hide it. Experiment with sample-precision delays and polarity/phase inversion, or an All Pass filter and just go with what sounds best.

But I admit for parallel compression, I prefer the compressors that have a wet/dry balance to avoid this problem.

[Berolzheimer]

And of course it's not only phase shift that will prevent nulling, it's any change to the sound- any type of distortion.

[scott petito]

HI Guys

thanks for the comments....I have long used analog compression for this application and been pretty happy with the results over the years.....ITB not so much.... I'm on an icon now so it's becoming kinda mandatory to get it to be a good thing since hardware and dealing with A/D delay is not so good

the whole thing again stemmed () from not trusting delay compensation in PT HD... just a feeling I was getting.....by the way the clue in was monitoring headphone mixes on a more me type box and finding a lot more annoying phase shift ( and yes I have delay compensation off for tracking and no plugs, this still bugs me though)

I realize that phase shift is a natural byproduct of any modification to the signal.. I was just surpised to see how different each type of compressor was and the fact that once I was aware of it it became very obvious to me.. much more so than any analog compression I've done....

but like anything else.. there is good phase shift and bad

I also began to wonder if it was bit rate reduction due to compression that was causing the "bad phase"
isn't that George Massenburg's pet peeve with plugin compression?

Anyway it made me rethink my application.... I will say that when I try to null the various compressors the ones that have the subjectively best sounding phase shift sound the best in the mix as well....

cheers
SP

[bdmctear]

Scott,

From reading the original post, it sounded like you came up with an interesting effect that you will use in the future. Can you clarify this? It sounds interesting, but I am not sure exactly what steps you took to arrive at the discovery, and what exactly it sounded like and how you might use it.

Brian

[nickynicknick]

yup

[Francis Vaughan]

Something to consider. Delay and phase are not the same thing. Delay compensation is needed to compensate for the number of samples needed in the plugin pipeline, but this is a fixed time period. Phase is an angle and is only meaningfully turned into a time period for one frequency.

An ideal compressor plugin should only cause a fixed delay. Which is what you have found. However a real compressor, and we would hope any plugin designed to emulate it, may include some frequency shaping. Any shaping will result in phase changes to the signal - that is how they must must, either real or emulated.

Where things get interesting is the case where a compressor (or indeed any effect) provides for pre and post effect shaping (this might not be under user control, merely part of the design). If the effect produced some harmonic distortion (which even an ideal compressor does, and real ones are prized for) the harmonics appear in-phase with their fundamental, but possibly with some lead or lag with respect to signal at the same frequency in the input. Thus a compressor that is apparently flat wrt frequency may actually create output components with unexpected phase lead and lags. Plugins that emulate real compressors, well one would really hope they would exhibit this.

But this is independent of the number of pipeline stages needed to perform the emulation, which is what delay compensation copes with. So indeed, what you describe is exactly what one would hope for. Further, if you did the experiment entirely OTB, the same result would be expected.

[contempo]

Originally Posted by Francis Vaughan

Something to consider. Delay and phase are not the same thing. Delay compensation is needed to compensate for the number of samples needed in the plugin pipeline, but this is a fixed time period. Phase is an angle and is only meaningfully turned into a time period for one frequency.

but delay and phase are indeed related, like cousins. i think i understand where you're coming from. not much of a physics wiz, i'll admit it!!! it's interesting though because a few years ago i tried making the move to an all ITB mixing setup and just couldn't make it work for me. lord knows it wasn't for lack of effort or lack of money to throw at it. at the end of the day i couldn't get past things sounding like they had a phaser stuck in the middle of a sweep and tearing my hair out trying to adjust this & that to make it work when all i wanted to do was make music. i ended up buying a little allen & heath board and rebuying some outboard gear, much better!

[Mike Caffrey]

Originally Posted by scott petito

I was experimenting today with plugins for parallel compression and eq for ITB...
what started out as a way to verify delay compensation turned into an interesting study of phase shift

I sent my 10 tracks of drums to 2 stereo auxes bus1-2 and bus 3-4
I made sure all levels were Identical to the sends by sending automation data to sends

then I put a trim plug on both aux tracks , I flipped the phase one aux bus trim bus1-2
I then paned both aux to center (mono)

the audio then cancelled 100% as it should.... I then decided to test delay compensation for Compression

as i inserted various compression plugs on aux 2 feed by 3-4 I noticed they all produced various phase shifts...remember the trims are still out of phase
even at unity gain with no compression some compressors "leaked audio" more than others

I tried tdm and rtas versions of the plugs some like most of the waves plugs sounded identical

the BF actually sounded quite different in RTAS and TDM

A UAD Fairchild was especially "leaky" at what I believed was unity settings...

remember I'm still only listening to the sum of the 2 out of phase buses

here's where it got interesting I tried to get a compression setting engaged and adjust it to produce as little leakage as possible( which obviously is phase shift caused by the change in level and attack release times) basically I was using attack and release times and some level compensation to get as close to phase cancelation as I could...

I then flipped the phase on the first bus... thus bringing all the drums back in to the mix including the parallel buss compression

it sounded really good....
here s the rub when I tried to increase the parallel buss fader to get more compression I became immediately aware of the phase shift ... it was subtle but obvious...

I tried this with various compressors at similar attack and release times and some compressors produced much more pleasant but still obvious phase shift
I started to wonder if this was due in part to the coloration of various compressors

the UAD Fairchild seemed to be the most forgiving and the dyn 3 was pretty good
others were hit and miss... strangely enough the linear multibands i used were the least forgiving
I did this with various eq's as well very interesting results there as well

well to end a long winded story .. I will be checking my parallel compression in mono and out of phase....

anyone else try this? ... I'm going to give it go with outboard but I already have convertor delay to cope with...


cheers
SP

I think this is a very cool test.

One thing that people do model is the phase shift of a device. Clearly you're aware that any alteration of sound, such as the attack an release time, will stop the null.

If you put an EQ on and started making changes, you'd cancel the null too. A very colored sounding plug in should cancel the null as well because it's changing the tone.

I'd be curious to know how it sounded if you did the same thing without delay compensation, but an identical plugin path, with one side muted. In theory, that should cancel the null too, but it might cancel it differently .

[scott petito]

Originally Posted by Mike Caffrey

I think this is a very cool test.

One thing that people do model is the phase shift of a device. Clearly you're aware that any alteration of sound, such as the attack an release time, will stop the null.

If you put an EQ on and started making changes, you'd cancel the null too. A very colored sounding plug in should cancel the null as well because it's changing the tone.

I'd be curious to know how it sounded if you did the same thing without delay compensation, but an identical plugin path, with one side muted. In theory, that should cancel the null too, but it might cancel it differently .

yup I didn't explain it as well I could but you hit the nail on the head.... I did try it with delay compensation on and off and identical plugin paths and to my ears it did phase in a different way obvious all subjective....

It all made me think that the subjective "quality" of Phase relationship might be a big reason people don't care for the sound of software compression.....

I'm all still intrigued by the idea that software compression create bit reduction especially at extreme setting. every 6 db of gain reduction lops of a bit or so and if you are doing this to a hi gain source it messes with resolution and image

oh well who know... something else to obsese about

cheers
SP

[seansolo]

OK. No ADC. 12 tracks of drums. Outputs (not sends, mind you) to bus 1-2. Outputs AGAIN to bus 3-4 (holding shift-ctrl-option) - to two auxes. Identical plugs on both, labeled 'drums' and 'squish'. You can prolly guess which one had the bypasses plugs on it. Notice my drums only go out A 1-2 on the output of the auxes. No problems. No phase issues. The only two plugs I really use on the auxes are the MDW5 band and the EMI/TG Chandler comp. Big, sick drum sound. No phase anamolies, both busses are being summed together. I also insert EQ's (also MDW's) on each channel - bypassed until needed. This makes it easier to add some EQ without throwing my whole kit out of phase. Everyone is happy. Why are you guys losing your mind on this one?

[contempo]

Originally Posted by scott petito

It all made me think that the subjective "quality" of Phase relationship might be a big reason people don't care for the sound of software compression.....

I'm all still intrigued by the idea that software compression create bit reduction especially at extreme setting. every 6 db of gain reduction lops of a bit or so and if you are doing this to a hi gain source it messes with resolution and image

huh. go figure, but that makes total sense. wouldn't we not have this problem, at least in theory by using plug-ins that run at 32bit or 64bit oversampling? i'd imagine that maybe there's still a reduction in bit depth but that with the 64bit processing, it wouldn't be noticeable to any living human... supersonic animal stuff. i don't mind plugs on individual tracks and use them quite a bit, but all the mults & final summing happens in analog or it gets weird & phasey.

[u b k]

whenever a computer sums two nearly identical versions of a given sound, it never sounds right to my ears. it's always phasing, the sound loses solidity, some part of the spectrum becomes hollowed, transients are weaker.


gregoire
del
ubk
.

[princeplanet]

Are you guys talking about issues when both the main and the parallel are summed to the same path? (eg, 1+2) or when they output to say 1+2 and 3+4 respectively and THEN sum to a common output such as 5+6?

I recall hearing a difference. Can anyone explain why?

[princeplanet]

Originally Posted by princeplanet

Are you guys talking about issues when both the main and the parallel are summed to the same path? (eg, 1+2) or when they output to say 1+2 and 3+4 respectively and THEN sum to a common output such as 5+6?

I recall hearing a difference. Can anyone explain why?

Anyone?